We revisited a classic study of morphological variation in the oldfield mouse (Peromyscus polionotus) to estimate the strength ofselection acting on pigmentation patterns and to identify the underlying genes. We measured 215 specimens collected by FrancisSumner in the 1920s from eight populations across a 155km, environmentally variable transect from the white sands of Florida’sGulf coast to the dark, loamy soil of southeastern Alabama. Like Sumner, we found significant variation among populations:mice inhabiting coastal sand dunes had larger feet, longer tails, and lighter pigmentation than inland populations. Most striking,all seven pigmentation traits examined showed a sharp decrease in reflectance about 55 km from the coast, with most of thephenotypic change occurring over less than 10 km. The largest change in soil reflectance occurred just south of this break inpigmentation. Geographic analysis of microsatellite markers shows little interpopulation differentiation, so the abrupt change inpigmentation is not associated with recent secondary contact or reduced gene flow between adjacent populations. Using thesegenetic data, we estimated that the strength of selection needed to maintain the observed distribution of pigment traits rangedfrom 0.0004 to 21%, depending on the trait and model used. We also examined changes in allele frequency of SNPs in twopigmentation genes, Mc1r and Agouti, and show that mutations in the cisregulatory region of Agouti may contribute to this clinein pigmentation. The concordance between environmental variation and pigmentation in the face of high levels of interpopulationgene flow strongly implies that natural selection is maintaining a steep cline in pigmentation and the genes underlying it.
O R I G I NA L A RT I C L E doi:10.1111/j.1558-5646.2008.00425.x NATURAL SELECTION ALONG AN ENVIRONMENTAL GRADIENT: A CLASSIC CLINE IN MOUSE PIGMENTATION Lynne M Mullen1,2 and Hopi E Hoekstra1,3 Department of Organismic and Evolutionary Biology and The Museum of Comparative Zoology, Harvard University, 26 Oxford Street, Cambridge, Massachusetts 02138 E-mail: lmullen@oeb.harvard.edu E-mail: hoekstra@oeb.harvard.edu Received January 30, 2008 Accepted April 22, 2008 We revisited a classic study of morphological variation in the oldfield mouse (Peromyscus polionotus) to estimate the strength of selection acting on pigmentation patterns and to identify the underlying genes We measured 215 specimens collected by Francis Sumner in the 1920s from eight populations across a 155-km, environmentally variable transect from the white sands of Florida’s Gulf coast to the dark, loamy soil of southeastern Alabama Like Sumner, we found significant variation among populations: mice inhabiting coastal sand dunes had larger feet, longer tails, and lighter pigmentation than inland populations Most striking, all seven pigmentation traits examined showed a sharp decrease in reflectance about 55 km from the coast, with most of the phenotypic change occurring over less than 10 km The largest change in soil reflectance occurred just south of this break in pigmentation Geographic analysis of microsatellite markers shows little interpopulation differentiation, so the abrupt change in pigmentation is not associated with recent secondary contact or reduced gene flow between adjacent populations Using these genetic data, we estimated that the strength of selection needed to maintain the observed distribution of pigment traits ranged from 0.0004 to 21%, depending on the trait and model used We also examined changes in allele frequency of SNPs in two pigmentation genes, Mc1r and Agouti, and show that mutations in the cis-regulatory region of Agouti may contribute to this cline in pigmentation The concordance between environmental variation and pigmentation in the face of high levels of interpopulation gene flow strongly implies that natural selection is maintaining a steep cline in pigmentation and the genes underlying it KEY WORDS: Adaptation, Agouti, color, Haldane, Mc1r, natural selection, Peromyscus Clinal variation can provide strong evidence for adaptation to different environments Although there are many examples of clines in allele frequencies (e.g., Berry and Kreitman 1993; Eanes 1999; Schmidt and Rand 2001; Fry et al 2008), often a precise understanding of the selective agent responsible for producing the cline is lacking On the other hand, there are many examples of clines in phenotypic traits (reviewed in Endler 1977; Hedrick 2006; Schemske and Bierzychudek 2007), but in most cases we not know their genetic basis Thus, there are few cases in which clinal variation in both phenotype and its underlying genes have been examined in an ecological context (but see Caicedo et al 1555 C 2008 The Author(s) Journal compilation Evolution 62-7: 1555–1570 C 2004; Hoekstra et al 2004) Here, we revisit a classic evolutionary study of deer mice (genus Peromyscus) conducted by Francis Bertody Sumner nearly a century ago to determine the role of natural selection in maintaining clinal variation in pigmentation and to identify the underlying genetics In the summer of 1924, after more than 20 years of studying geographic variation in Peromyscus in the western United States, Francis Sumner traveled to Florida, “where material has been found which promises to yield more valuable returns than any which have thus far been reported” (Sumner 1926, p 149) Sumner was referring to the widespread species Peromyscus polionotus, 2008 The Society for the Study of Evolution L M MULLEN AND H E HOEKSTRA which, throughout most of its range in the southeastern US, has a dark brown dorsal coat, striped tail, and light gray ventrum A colleague, however, had recently reported the existence of a unique race, P p leucocephalus, on a barrier island off the Gulf Coast of Florida (Howell 1920)—a race with “paler coloration and more extensive white areas than any other wild mouse with which [Sumner was] acquainted” (Sumner 1929a, p 110) During 1926 and 1927, Sumner collected, measured, and prepared museum skins for over 400 mice from eight localities spanning over 150 km from a coastal population in Florida northward into Alabama As Sumner found, P polionotus occupied an environmental gradient ranging from the vegetatively depauperate, brilliant white sandy beaches of Florida’s Gulf coast to inland areas, particularly densely vegetated oldfields Sumner’s two-year effort produced not only laboratory stocks derived from the two extreme forms (taken from end points of his transect), but also an extensive series of museum skins In the laboratory, crosses among light and dark forms convinced Sumner that (1) the difference in coat color between the beach and inland form was genetic and controlled by a handful of loci, and (2) the two extreme forms were interfertile and thus could exchange genetic material in the wild (Sumner 1930) Sumner also used his study skins to examine variation in nature He found that, about 50 km from the coast, the pale-colored mouse was replaced rather abruptly by a darker form more typical of the genus—a narrow “area of intergradation,” (Sumner 1929a,b) Taken together, these observations were puzzling: why was there such an abrupt change in pigmentation if the mice were interfertile? Several answers have been offered Sumner speculated that these distinct racial differences were largely caused by the range expansion of light mice into the previously allopatric range of dark mice (Sumner 1929b) Bowen (1968) also believed that the two divergent forms had recently come into contact, but he emphasized that ecological barriers were necessary to limit intermixing between them and that selection need not be invoked to explain the observed cline Still others favored the idea that strong disruptive selection was imposed by a sharp break in soil color, so that mismatched hybrid offspring had low survival (Haldane 1948) This last explanation gained favor, and Sumner’s work on this transect became a classic textbook example of adaptation, widely discussed in the evolutionary literature (Dobzhansky 1937; Mayr 1942, 1954, 1963; Huxley 1943; Haldane 1948; Ford 1954, 1960, 1964) The pervasive interest in this example comes from its instantiation of selection in action In his now-famous 1948 paper, “The theory of a cline,” J B S Haldane used Sumner’s transect as an example of how to estimate selection coefficients from geographic variation in allele frequencies (Haldane analyzed one trait, the cheek patch, whose variation, he assumed, was caused by allelic differences at a single locus) Haldane concluded that selection on cheek patch 1556 EVOLUTION JULY 2008 color was strong, but added that this conclusion relied on several untested assumptions about, for example, migration rates and the underlying genetics By returning to this classic system armed with modern molecular tools and more sophisticated mathematical methods, we can gain new insight into how the interaction between selection, demography, and genetics generates this striking cline in pigmentation Here, we tested the three alternative hypotheses about the genetic structure of Sumner’s mouse populations and showed that habitat-specific natural selection is responsible for clinal variation in pigmentation We also estimated the strength of selection acting on pigmentation by supplying the missing genetic data called for by Haldane Finally, we determined geographic changes in allele frequencies of candidate pigmentation genes in an effort to identify genes and mutations responsible for this cline in coat color All these data were gathered using Sumner’s original museum collections of P polionotus Methods and Materials MOUSE SAMPLES We sampled 215 museum specimens caught by Sumner in 1927 and now accessioned at the University of California-Berkeley Museum of Vertebrate Zoology These mice were collected along a southwest-to-northeast transect spanning 155 km from St Andrews Bay, Florida to southeastern Alabama (Fig 1; Sumner 1929) Table provides detailed locality data We characterized morphological and genetic diversity in 22–25 samples from each of eight sampling locales with one exception: we sampled 43 individuals from one “population” [Intergrades (IN)] The Intergrades site comprises four localities no more than km apart, and these localities span the heterogeneity in soil color at the center of the transect Accession numbers for each mouse are listed by collecting site in online Supplementary Table S1 Geographic distances between collection sites were calculated with the GEOD (http://www.indo.com/distance/) program from the U.S Geological Survey Because the sampling transect is not perfectly linear, we used measures of latitude as a proxy for actual distance between the populations The average distance between collecting sites was 15.5 km (± 22.0 km), which is larger than the estimated dispersal distance for these mice (Blair 1951; Swilling and Wooten 2002), and therefore each population can be considered independent LINEAR MEASUREMENTS To quantify morphological variation, we recorded four standard museum measurements (to the nearest 0.5 mm) for each specimen: hindfoot length (distance from heel to middle digit), tail length (base to tip), ear length (ear base to the most distant tip), and body length (from tip of the rostrum to the base of the tail) Online Supplementary Table S2 gives trait means for mice from each of the C L I NA L S E L E C T I O N I N P E RO M Y S C U S Figure Map showing the 155-km transect from northwestern Florida to southeastern Alabama; collection sites are indicated by colored circles and listed by abbreviation (see Table 1) Representative mice from five of the eight populations are shown with a soil sample from each site Multiple mice and soil samples are shown from the Intergrade (IN) population to highlight heterogeneity at this locale Additional soil samples were taken in the primary transect (circles) as well as in two parallel transects to the west (squares) and east (triangles) eight collecting sites We did not find any statistical differences between our measurements and Sumner’s original measurements for these samples We also included the body weight of the mice from Sumner’s records on the live mice We used analyses of variance (ANOVAs) to summarize the variation in body dimensions All statistical and multivariate analyses were performed using JMP version 5.1.2 (SAS Institute, Cary, NC) SPECTROPHOTOMETER MEASUREMENTS To measure color, we used a USB2000 (Ocean Optics, Dunedin, FL) spectrophotometer with a dual deuterium and halogen light source to capture spectral reflectance from museum skins and surface soil samples (see below) The probe was held at a 45◦ angle to the surface parallel to the body axis, and the program OOIBASE32 (Ocean Optics) was used to capture the wavelength readings We recorded reflectance of three facial traits (rostrum, whisker, and cheek patch) and three body traits (dorsal stripe, flank, and ventrum); these traits show the most divergence in pigmentation among populations (following Hoekstra et al 2006) For each trait, reflectance was measured in triplicate using the visible spectrum of most visual predators (300–700 nm; Bennett and Cuthill 1994) We used the segmental classification method to quantify three aspects of color: brightness, chroma, and hue (following Endler EVOLUTION JULY 2008 1557 L M MULLEN AND H E HOEKSTRA Table ∗ Collection sites with abbreviations, sample sizes (N), distance from the Gulf Coast of Florida, and geographic coordinates Collection site Abbreviations N Distance from Gulf Coast (km) Latitude (N)/ Longitude (W) Western Peninsula Crystal Lake Round Lake Intergrades∗ Alford Kent Mill Oak Hill Orange Hill Chipley Dekle Plantation Graceville Abbeville WP CL RL IN 25 25 25 43 17 16 25 22 25 25 31 50 30.2200/−85.8809 30.4441/−85.6885 30.6516/−85.3888 55 56 58 59 64 72 83 155 30.6947/−85.3931 30.6638/−85.4319 30.7113/−85.4716 30.6832/−85.5642 30.7819/−85.5285 30.8522/−85.3494 30.9569/−85.5166 31.5663/−85.2513 CH DP GR AB Intergrades “population” contain four clustered collecting sites 1990) We averaged reflectance readings every nm, sorted the reflectance data into four bins of 100 nm each (A = 300 to 400 nm; B = 400 to 500 nm; C = 500 to 600 nm, and D = 600 to 700 nm), and then calculated the total brightness in each bin These reflectance data were also used to calculate chroma ((D − √ B)2 − (C − A)2 ) and hue (arcsine (C − A)/ chroma) As above, we analyzed reflectance of each individual’s raw data and then calculated the average reflectance for each collecting site (online Supplementary Table S3) Brightness values were converted to percent reflectance relative to a white standard To maximize differences among populations, we normalized the population means so that the darkest population for a particular trait was assigned a value of 0, whereas the lightest population was given a value of In addition to spectrophotometric measurements, we also measured tail pigmentation by recording the percent of the tail that had a visible pigment stripe SOIL SAMPLES To document variation in substrate color, we collected surface soil samples along three transects (Fig 1) First, we collected soil at each of the eight sites where Sumner caught mice At each of Sumner’s sites, 10 soil samples, spaced approximately 10 m apart, were collected In addition, more fine-scale sampling was done at shorter intervals: one soil sample was collected every km in between Sumner’s collection sites Finally, we conducted two parallel transects roughly 25 km to the west and east of Sumner’s transect, where samples were also collected at km intervals Spectrophotometric measurements were taken as described above (online Supplementary Table S3) DNA EXTRACTIONS We extracted DNA from all 215 specimens using a mm × mm piece of dried museum pelt cut from the ventral surface of 1558 EVOLUTION JULY 2008 each sample We took several steps to avoid contamination during DNA extraction First, we sterilized scissors between samples Extraction and PCR steps were done in a laboratory free from vertebrate PCR products Extractions were performed in small groups (no more than 10 individuals per session), which included individuals from only one population Finally, negative extraction and PCR controls were always included to detect contamination Each sample was shaved with a razor blade to remove the fur and then rinsed in 100% ethanol Ethanol washes of the skin sample were performed roughly every h over a 24-h period to remove salts and downstream PCR inhibitors Following these washes, we extracted DNA using the DNeasy Tissue Extraction kit (Qiagen, Valencia, CA) with the following modifications: (1) AE Buffer was diluted 1:10 in Milli-Q grade H O and heated to 70[◦ C, (2) 50 μl of AE Buffer was applied to the spin columns for the first elution step, and repeated using 100 μL, and (3) following each application of Buffer AE, spin columns were incubated at RT for before the final elution Extractions from all samples were stored at −80◦ C MICROSATELLITE REACTIONS PCR reactions were performed for 10 microsatellite markers: Bw2–25, Bw3–12, Bw4–5, Bw4–13, Bw4–110, Bw4-SREL, Bw4– 137, Bw4–234, Po-17, and Po-26 (see Mullen et al 2006 for primer sequences and reaction conditions) We made the following modifications to the reactions because the DNA samples were presumed to be degraded: (1) the master mix included 1.5 μl of 10× Bovine Serum Albumin (BSA) and an additional 1.5 μl of 25 mM MgCl , (2) annealing times were increased to 45 sec, and (3) 60 cycles were performed for all markers All microsatellites were scored on an ABI 3100 sequencer using a ROX 400HD ladder, and results analyzed using the GENEMAPPER version 3.5 software (Applied Biosystems, Foster City, C L I NA L S E L E C T I O N I N P E RO M Y S C U S CA) Because null alleles and allelic dropout are often seen in studies that use highly degraded DNA specimens (Taberlet et al 1996; Bonin et al 2004), we used several strategies to closely monitor results from microsatellite markers First, we ran three replicates of all markers for a subset of 20 individuals to judge the error rate for each marker, which we found to be low (< 5%) The amount of DNA present in all samples was also measured with a SYBR green quantitative PCR reaction, using marker Bw4–137 The minimum threshold for a sample’s use was set at 25 pg, as lower values tend to yield high rates of allelic dropout (Morin et al 2001) We calculated observed and expected heterozygosity as well as the number of alleles using the program GENALEX version (Peakall and Smouse 2006) Deviations from Hardy– Weinberg equilibrium were calculated using exact tests implemented in GENEPOP version 3.4 (Raymond and Rousset 1995) to identify markers with a deficiency in heterozygote genotypes To determine whether variation in pigmentation is best explained by geographic or genetic distance, we examined statistical associations among three matrices: (1) variation in pigmentation among populations, (2) genetic differentiation, and (3) geographic distance, by performing Matrix Correspondence Tests (MCTs) using LAPROGICIEL version R 4.0 (Casgrain and Legendre 2001) MCTs use a regression design to test for multiple correlations between matrices and use permutation methods to assess statistical significance We performed pairwise MCTs to test for correlation between relative brightness values averaged over all traits (pigmentation), geographical distance (geography), and genetic distance (linearized F ST ) Partial MCTs were used to detect correlations between two variables (pigmentation and geography, or pigmentation and genetic distance) while controlling for the effect of the third (either genetic distance or geography, respectively) ANALYSIS OF CLINE SHAPE ANALYSIS OF POPULATION STRUCTURE Data from all 10 microsatellite markers were used to estimate levels of gene flow and genetic differentiation between adjacent populations and among all populations To determine the degree of genetic structure, we calculated F ST using the program ARLEQUIN version 2.000 (Schneider et al 2000) We also calculated R ST values, which are based on a stepwise mutation model most appropriate for microsatellites (Slatkin 1995) Because R ST values did not differ statistically from F ST , we report only F ST values here To test for a pattern of isolation by distance, we regressed linearized F ST [F ST /(1 − F ST )] against distance (km) between sampling locales Maximum-likelihood (ML) estimates of the pergeneration effective number of migrants (Nm) between pairwise populations were calculated with the program MIGRATE version 2.0.3 (Beerli and Felsenstein 1999) We initially used MIGRATE default parameters, and then optimized the likelihood score by using values from the previous run as the next run’s starting parameters until the ln (L) peaked and stabilized To further characterize the extent of genetic structure among populations, we used the Bayesian clustering program STRUCTURE version 2.0 (Pritchard et al 2000) to probabilistically assign individuals to subpopulations (k) independently of sampling areas We first tested the hypothesis that two previously isolated (and genetically distinct) populations have recently come into contact by constraining our analyses to k = We then separately calculated the maximum-likelihood estimate of the number of subpopulations using the admixture-with-uncorrelated-allelefrequencies model with a burn-in of 30,000 Markov Chain Monte Carlo (MCMC) steps followed by 100,000 iterations The highest value of k from the average of 10 runs for k = – 11 (where the highest exploratory k was determined by using the number of populations sampled plus three) was confirmed with the k algorithm (Evanno et al 2005) To mathematically describe clinal variation, we generated maximum-likelihood curves for each pigmentation phenotype using the one-dimensional transect option in the program ANALYSE version 1.30 (Barton and Baird 1995), which fits sigmoidal curves to cline data and searches for the maximum-likelihood estimates of cline parameters using a Metropolis Hastings algorithm (Szymura and Barton 1986) We allowed four parameters to vary for each trait: p , p max (minimum and maximum frequencies at the tails of the cline, respectively), cline center, and cline width Cline center is the distance between the geographic position where the maximum slope occurs and the first (beach) population For phenotypic traits, cline width is calculated using the equation w = z/(∂z/∂t), where z is the difference in population means on each side of the cline and ∂z/∂t is the maximum slope (Barton and Gale 1993) For allele frequencies, cline width is calculated using the inverse of the maximum slope Because extensive sampling was not performed at the ends of the cline, we did not explore the stepped-cline model Body measurements were not analyzed using this approach, as they did not vary clinally To test for statistical concordance among clines (for both phenotypic traits and allele frequencies), we used likelihood-ratio tests to compare cline centers and widths (per Phillips et al 2004) We constructed likelihood profiles by summing the highest loglikelihood scores generated by ANALYSE for all phenotypes to yield an ML estimate of a cline parameter shared by all traits We subtracted these profiles from the sum of the ML scores for each trait and used likelihood ratio tests to determine if cline centers (or widths) were statistically different The likelihood test follows a χ2 distribution, in which the degrees of freedom are equal to the difference in the number of parameters between each model (in this case, the number of phenotype — 1) For both cline center and width estimates, we calculated the ± log L range, which is equivalent to 95% confidence intervals (Edwards 1992) EVOLUTION JULY 2008 1559 L M MULLEN AND H E HOEKSTRA ESTIMATING SELECTION COEFFICIENTS We used two methods to estimate the strength of selection acting on pigmentation We first used a model that assumes the environment changes gradually across the transect, in which dispersal distance is much smaller than the cline width (the gradient model) This model assumes the equation b = l2 (2.40/w)3 , in which w is the width of the cline, l is the standard deviation of the adult-offspring dispersal distance, and b represents the selection gradient (Endler 1977) We also used a model that assumes a step-like change in the environment, in which dispersal distance is similar to or larger than the cline width (the ecotone model) √ This model uses the equation w = σ/ s, in which w is the width of the cline, σ is the standard deviation of the adult-offspring dispersal distance, and s is the selection coefficient (Haldane 1948; Bazykin 1969; Slatkin 1973) We did not directly estimate dispersal distance in these populations, and instead used an estimate of subadult dispersal (0.260 km) taken from mark–recapture studies of P polionotus (Blair 1946; Swilling and Wooten 2002) ALLELIC VARIATION IN CANDIDATE PIGMENTATION GENES To identify genetic regions that may underlie clinal variation in pigmentation, we focused on two candidate genes Previous work using a genetic mapping approach identified the melanocortin1 receptor (Mc1r) and the Agouti signaling protein (Agouti), as major contributors to the phenotypic differences between two related subspecies, the light-colored Santa Rosa Island beach mouse (P p leucocephalus) and a darker mainland subspecies (P p subgriseus; Steiner et al 2007) Although these subspecies are different from those studied here (and occur in different areas), they show a similar divergence of phenotypes To determine if allelic variation at Mc1r and Agouti contributes to clinal variation in pigmentation, we measured allele frequencies within and between populations For Mc1r, we genotyped a single nucleotide mutation in the coding region, which codes for an amino acid change known to decrease the receptor activity in laboratory experiments (Hoekstra et al 2006) For Agouti, expression differences (as opposed to amino acid change) are correlated with variation in pigmentation, but the causal mutation(s) have not yet been identified (Steiner et al 2007) Therefore, we genotyped four single nucleotide polymorphisms (SNPs) that span 100 kb of the Agouti regulatory region (approximately kb, 40 kb, 50 kb and 100 kb of exon 2), as these may be in linkage disequilibrium (LD) with a causal mutation We used alignments of several vertebrate Agouti sequences in GenBank to design conserved primers and amplify about kb regions in Peromyscus In these regions, we identified SNPs that were fixed differences between three mainland (P p subgriseus) and three beach mice (P p leucocephalus) from which we had fresh, nondegraded DNA samples For all SNPs, we define the “dark” allele 1560 EVOLUTION JULY 2008 as the allele present in the mainland population and the “light” allele as the allele present in the beach population For four of the five SNPs, we designed custom TaqMan SNP assays (Applied Biosystems) Online Supplementary Table S4 provides the primer and probe sequences We conducted all reactions in 13 μl reaction volumes containing μL DNA, 6.25 μl 2X TaqMan Universal PCR Master Mix (Applied Biosystems), 1.5 μl 10X BSA, and 0.3125 μl custom probe We genotyped SNPs on an ABI Prism 7000 using an initial AmpErase uracil-N-glycosylase (UNG) activation step of 50◦ C for min, a denaturation step of 95◦ C for 10 followed by 40 cycles of 94◦ C for 15 sec and 60◦ C for We scored SNPs with the Allelic Discrimination Assay procedure in SDS version 1.1 (Applied Biosystems) For one Agouti SNP, located 100 kb upstream of exon 2, we could not design a TaqMan probe Therefore, we PCRamplified and sequenced 235 bps using primers Ag5 F (5 AACTTGCTTTGTAGACC ) and Ag5 R (5 TGGGGGAATCCAACCTG ) We performed PCR reactions in a 15 μl volume with μl of template DNA, 10× Taq Buffer (1.5mM MgCl ; Eppendorf), 1.5 μl of 25 mM MgCl , 1.5 μl 10× BSA, 0.3 μl of 10 μM dNTPs, 0.6 μL of each 10 μM primer, and 0.2 U of Taq DNA Polymerase (Eppendorf, Westbury, NY) Reaction conditions were as follows: an initial denaturation step of 94◦ C for min, followed by 60 cycles of denaturation at 94◦ C for 30 sec, annealing at 58.8◦ C for 45 sec and extension at 72◦ C for We performed a final extension step at 72 ◦ C for 10 We purified PCR products by adding 0.08 μl Exonuclease I and 0.4 μl Shrimp Alkaline Phosphatase (USB Corporation, Cleveland, OH) in a total volume of μl to each sample Samples were incubated at 37◦ C for 15 then heated to 80◦ C for 15 We sequenced the products with BigDye version 3.1 (Applied Biosystems) using PCR primers and ran them on an ABI 3100 Genetic Analyzer We visualized and aligned sequences by eye using SEQUENCHER version 4.1 (Gene Codes Corporation, Ann Arbor, MI) ANALYSIS OF LINKAGE DISEQUILIBRIUM BETWEEN SNPs Because the phenotypic effects of Mc1r are dependent on the Agouti genetic background (Steiner et al 2007), we used the program TASSEL version 2.1 (Bradbury et al 2007) to estimate linkage disequilibrium (LD) between all pairwise combinations of Mc1r and Agouti SNPs We used two ways to test for nonrandom associations between SNPs First, r2 values estimate the correlation between two loci Since r2 values depend on allele frequencies (Hedrick 1987), we also calculated the normalized disequilibrium coefficient, D , which is directly related to the extent of recombination between two loci Both r2 and D values range from to 1, where represents linkage equilibrium and indicates complete linkage disequilibrium We ran 10,000 C L I NA L S E L E C T I O N I N P E RO M Y S C U S permutations and used a two-sided Fisher’s exact test to calculate significance ASSOCIATION BETWEEN GENOTYPE AND PHENOTYPE For each of the five SNPs, we assigned individuals to one of three genotypic classes, comprising homozygotes for the light allele, homozygotes for the dark allele, or heterozygotes We first tested for a statistical association between genotype and each of the seven pigmentation traits using Kruskal–Wallis nonparametric tests Because of the potential interaction between loci, we also tested for an association between genotypic class, when genotypes at both Mc1r and Agouti were considered simultaneously (a total of nine genotypic classes), and each pigmentation trait ENVIRONMENTAL VARIATION Several patterns emerged from the spectrophotometric measurements of soil collected along the three parallel transects (Fig 2C) First, for all three transects, the brightest soil occurred at the southernmost site (WP) on sandy coastal dunes Second, the northernmost populations had the darkest soil, and the soil is consistently dark Finally, the central IN site showed the most variation in soil brightness among transects, a pattern mirrored in the high variation in mouse coat color in that location However, we did not find evidence for an abrupt change in soil color at the IN site, as hypothesized by Sumner (1929b) Thus, although we found significant variation in soil color, with lighter soil in the south compared to darker soil in the north, intermediate sites showed a large degree of heterogeneity in brightness CORRELATION BETWEEN PHENOTYPE Results MORPHOLOGICAL VARIATION To determine the relationship between morphology and geographic distance, we plotted average values for both body dimension and pigmentation traits for each of eight populations against distances and conducted multivariate ANOVAs (Fig 2) Because males and females did not differ statistically for any traits, we combined the sexes in our analysis For two body traits we found significant variation among populations, foot length (F 1,203 = 4.31, P = 0.031) and tail length (F 1,212 = 3.79, P = 0.030), as well as a correlation between these traits and distance (Fig 2A) This pattern is driven entirely by the extreme phenotypes in the beach population—most of the change in foot length (> 1.5 mm) and tail length (> 5.5 mm) occurs between the coastal site (WP) and first inland site (CL), – 30 km from the coast Although both total body length (F 1,212 = 8.33, P = 0.004) and ear length (F 7,198 = 17.40, P < 0.001) varied among populations, these traits did not show a consistent directional change with distance Body weight did not vary significantly among populations or show a correlation with distance In addition to body measurements, reflectance of all seven pigmentation traits varied among populations (F 7,208 values ranged from 28 to 63, P < 0.0001) and with geographic distance (Fig 2B) Across the three most coastal populations, some traits were uniformly bright (e.g., cheek and dorsal stripe), whereas others (e.g., rostrum and flank) were increasingly dark with distance from the coast However, all pigmentation traits showed a steep cline in reflectance approximately 50 – 60 km from the coast (Fig 2B) And, in general, mice were fully pigmented north of this transition Analysis of two other color measures, hue and chroma, did not show clinal variation for any traits, because most of the change in mouse color (from white to brown pelage) is best captured by brightness AND ENVIRONMENT For all seven pigmentation traits, we found a significant association between the reflectance of pelage and the reflectance of the soil across the eight collection sites Regression coefficients (r2 ) ranged between 0.39 and 0.68 and all were significant at the P < 0.001 level; it is important to note that the phenotypic traits are genetically correlated with one another (Steiner et al 2007) and thus, their association with environmental variables is not independent We did not, however, observe an abrupt change in soil color at the precise location where we observed a sharp change in mouse pigmentation Most of the change in brightness of the soil occurred between the two southern sites, WP and CL, whereas most of the change in pigmentation occurred at or around the central IN population, which is 20 km north of the change in soil brightness Thus, although there are strong correlations between soil and mouse color, the break in mouse pigmentation and soil color is not perfectly coincident and thus requires explanation POPULATION STRUCTURE AND GENE FLOW All microsatellite loci were in Hardy-Weinberg equilibrium with the exception of two (Bw2–25 and Bw4–13), which were deficient in heterozygotes in all populations except the beach population Because heterozygote deficiency may indicate loci prone to allelic dropout (Taberlet et al 1996), we excluded these markers from further analyses The remaining eight microsatellite loci were all variable: the number of alleles ranged from – 15 (mean = 6.7), H O ranged from 0.200 to 1.00 (mean = 0.577), and H E ranged from 0.183 to 0.872 (mean = 0.641) across loci (online Supplementary Table S5) Pairwise estimates of population structure (F ST ) and gene flow (Nm) are shown in Table Populations showed a barely significant pattern of isolation-by-distance (r2 = 0.67, P = 0.04) However, overall estimates of F ST were low—the largest estimate was between the two terminal populations of the transect (WP and EVOLUTION JULY 2008 1561 L M MULLEN AND H E HOEKSTRA Figure Graphs showing change in body dimensions, pigmentation traits, and soil coloration across eight collecting sites Gray bars represent the “area of intergradation.” r and P-values from ANOVAs are given (A) Means (± SE) of three body dimensions (foot size, tail length, and body size) for each population (B) Cartoon highlighting six pigmentation traits Mean (± SE) of the relative brightness (percent relative reflectance) for each trait as well as the mean percentage of the tail that is striped is shown (C) Mean (± SE) of the relative brightness of soil collected every km along the main transect (circles connected by a solid line) as well as the western transect (squares with dashed lines) and eastern transect (triangles with dashed lines) shown in Figure Relative brightness was not correlated with geographic distance in any soil transect 1562 EVOLUTION JULY 2008 C L I NA L S E L E C T I O N I N P E RO M Y S C U S Table Pairwise estimates of population structure based on eight microsatellite markers surveyed in eight populations Esti- mates of F ST between sites (below the diagonal) and ML estimates of the effective number of migrants (Nm; above the diagonal) Estimates representing neighboring populations are in bold WP CL RL IN CH DP GR AB ∗ WP CL RL IN CH DP GR AB – 0.170 0.132 0.085 0.152 0.134 0.118 0.371 2.41 – 0.084∗ 0.069∗ 0.067∗ 0.019 0.020∗ 0.106 1.92 1.62 – 0.070 0.112 0.122 0.061∗ 0.188 2.12 1.74 5.80 – 0.035 0.051 0.030 0.153 2.27 1.12 1.48 3.25 – 0.025 0.016 0.163 2.19 1.20 1.63 2.03 6.75 – 0.022 0.141 1.84 3.48 5.47 3.51 2.52 4.31 – 0.105 3.45 2.98 8.56 3.74 2.70 4.68 3.34 – F ST values statistically indistinguishable from zero (P < 0.05) AB; F ST = 0.371) that are separated by 155 km Comparisons of neighboring populations also showed little evidence of genetic structure: all of the F ST estimates were low (F ST < 0.1), with the exception of the two terminal populations and their nearest neighbor (WP-CL, F ST = 0.170 and GR-AB, F ST = 0.105) In fact, the estimates of F ST between the central populations (IN to CH), where most of the change in pigmentation occurs, were among the lowest of all pairwise comparisons Overall, migration rates (Nm) were high, reflecting this low level of genetic structure (Table 2) The estimated number of migrants ranged from to individuals per generation between neighboring populations We also found high levels of admixture between populations When STRUCTURE was constrained to k = (lnP = −7029.5), we did not find any clustering of genotypes between the two subpopulations (Fig 3A) We found, instead, that the most likely number of populations is (lnP = −6272.9) Most individuals had equal membership among the clusters, with the exception of the coastal WP population, where 80% of genotypes from these samples were assigned to one cluster (Fig 3B) We averaged the highest probability of assignment (q) from each individual and found q = 0.49 ± 0.14 The probability of assigning any individual to its correct population is thus less than 50%, which again shows that extensive gene flow has homogenized subpopulations for these markers COMPARING PHENOTYPIC AND GENETIC VARIATION Using pairwise and partial MCTs, we found a highly significant correlation between pigmentation and geographic distance (Table 3; r2 = 0.574, P < 0.001) There was also a significant correlation between geographic and genetic distance (P = 0.001), but when we controlled for genetic distance, the correlation between pigmentation and geography remained (r2 = 0.528, P = 0.002) Moreover, although there was a correlation between pigmentation and genetic distance, when we controlled for the effects of geography, this correlation became nonsignificant Thus, variation in pigmentation is best explained by geographic distance, not genetic distance Graphical display of individual coefficients of membership in subgroups based on multilocus genotypes using the program STRUCTURE Each individual is represented by a vertical line, broken into k colored segments proportional to its membership in each of the Figure k subgroups Individuals are grouped by collecting site and separated by a vertical bar (A) Analysis constrained to k = subgroups (B) Analysis using the maximum-likelihood estimate of k = subgroups EVOLUTION JULY 2008 1563 L M MULLEN AND H E HOEKSTRA Table Pairwise and partial matrix correspondence tests (MCT) between genetic distance (pairwise linearized F ST ), geographic distance (km) and phenotypic variation (average brightness of seven pigmentation traits) Parentheses indicate variable whose effects were controlled for in the partial MCTs Regression coefficients (r ) and P-values are given Significance levels for all comparisons were generated from 10,000 permutations r2 F ST -Geography Pigmentation-Geography Pigmentation-F ST Pigmentation-Geography_(F ST ) Pigmentation-F ST _(Geography) ∗ 0.666 0.574 0.298 0.528 0.137 P-value 0.001∗ 48 km) Table Using the width of these phenotypic clines, we estimated the strength of selection acting on each trait (Table 4) Overall, selection coefficients were higher under the assumptions of the ecotone model than the gradient model Whisker and dorsal stripe had the highest estimates of selection (s = 0.4% or 21%, depending on the model), whereas the selection estimates for rostrum and ventrum were lowest (s = 0.0004% or 7%, depending on the model) VARIATION IN ALLELE FREQUENCY AT Mc1r AND AGOUTI We found distinct changes in allele frequency at each SNP in the two candidate pigmentation genes (Fig 4) For Mc1r, we found that the derived “light” Mc1r allele is present in seven of the eight populations up to a frequency of 0.34 The light allele was at intermediate frequency in both the dark- and light-colored terminal populations (AB = 0.13 and WP = 0.31) and absent in a central population (CH) For two Agouti SNPs, the change in allele frequency was not clinal For one of these, the SNP closest to the coding region (2 kb SNP), the light allele ranged in frequency from 0.71 to 1.00, reaching the lowest frequencies in the two terminal populations and fixation in a central population (CH) The light allele of the Agouti 50 kb SNP was either fixed or at high frequency in all populations, with the lowest frequency found in the GR population (0.95); therefore this marker was uninformative In contrast, for the two other Agouti SNPs (40 and 100 kb), allele frequencies showed clinal variation For both of these SNPs, the derived light allele was fixed in the three southernmost populations, then the light allele began to decrease in frequency starting in the central IN population, and, finally the light allele showed its lowest frequency in the northernmost population (0.62 for both SNPs) This pattern was most striking for the Agouti 40 kb SNP Moreover, using ANALYSE to fit a cline to the mean light allele frequency, we found that estimates of cline center and width for Estimates of cline center (km), cline width (km), and selection coefficient for seven pigmentation traits Maximum-likelihood curves were fitted to the average relative brightness for each population Cline center is shown in kilometers from the Gulf Coast Trait Center (±2 logL) Width (±2 logL) Selection coefficient ecotone model1 Selection coefficient gradient model1 Rostrum Whisker Cheek Dorsal stripe Flank Ventrum Tail stripe 52 (46–59)3 58 (56–63)2 58 (56–60)2 54 (52–55)3 52 (41–54)3 50 (44–56)3 52 (50–53)3 48 (22–67)5 (1–17)4 11 (1–17)4 (1–11)4 (4–51)4 60 (44–85)5 10 (7–20)4 0.07 (0.11–0.06) 0.21 (0.51–0.12) 0.15 (0.51–0.12) 0.21 (0.51–0.15) 0.17 (0.25–0.07) 0.07 (0.08–0.05) 0.16 (0.19–0.11) 8.5×10−6 (8.8×10−5 –3.1×10−6 ) 0.004 (0.93–0.0009) 0.0007 (0.93–0.0002) 0.004 (0.93–0.0007) 0.001 (0.01–7.0×10−6 ) 4.3×10−6 (1.1×10−5 –1.5×10−7 ) 0.001 (0.003–0.0001) Range of selection estimates are based on the range of cline width estimates (±2 logL) Cline center estimates that are statistically equivalent for two traits Five traits for which cline center estimates are statistically equivalent Cline width estimates that are statistically concordant for five traits Traits for which width estimates are statistically concordant with each other Two different cline centers and two separate cline widths were identified 1564 EVOLUTION JULY 2008 C L I NA L S E L E C T I O N I N P E RO M Y S C U S Figure Frequency of five SNPs in two pigmentation genes, Mc1r and Agouti, among 10 populations in the southeastern US Map shows eight collecting sites along the transect (circles) and two additional collecting sites: Santa Rosa Island, Florida and Lake Louisa State Park, Florida (squares) The genomic structure of Mc1r and Agouti (based on the Mus structure) is shown; large gray boxes represent coding exons (exons – 4), and small boxes represent untranslated exons (exons 1a, 1a’, 1b, and 1c) Stars show the approximate location of each SNP Pie diagrams represent the genotype frequencies in each population: alleles identical to light-colored mice from Santa Rosa Island are shown in white and those identical to dark-colored mice from Lake Louisa are shown in black only the 40 kb SNP were statistically concordant with phenotypic clines For example, for the 40 kb SNP, we estimated the cline center to be at 56 km (± 2logL; 53–62 km) from the coast and the width to be km (± 2logL; 2–29 km) This cline in allele frequency was statistically indistinguishable from all seven pigment traits (χ2 = 8.52, df = 6, P = 0.202) for cline center and five traits (whisker, cheek, flank, dorsal stripe, and tail stripe; χ2 = 3.60, df = 4, P = 0.608) for cline width Even for the 40 kb SNP, however, the association between allele frequency and phenotype clines was not perfect For example, the dark allele was never fixed in any of the three northernmost populations that harbor the darkest mice: the dark Mc1r allele reached 87.5% and the 40 kb Agouti SNP reached only 38.0% in the terminal population LINKAGE DISEQUILIBRIUM BETWEEN whereas D ranged from 0.063 to Only two pairs of SNPs showed a significant departure from linkage equilibrium Two neighboring SNPs in Agouti, located 40 kb and 50 kb upstream of exon 2, had significant departures from linkage equilibrium This may reflect their physical proximity—10 kb apart—or may be only a biased estimate of LD due to the lack of variation at the 50 kb SNP Strikingly, the Mc1r SNP and the 40 kb Agouti SNP also showed a significant LD for both measures (r2 = 0.021, D = 0.425) GENOTYPE–PHENOTYPE ASSOCIATIONS When all the individuals were considered together, we found a significant association between some genotypes and pigmentation (Table 6) First, we found a significant correlation between Mc1r and pigmentation for four traits (dorsal stripe, flank, ventrum and Mc1r AND AGOUTI SNPs All pairwise estimates of LD between Mc1r and Agouti SNPs are reported in Table r2 values ranged from 0.001 to 0.028, Table Association (χ2 values) between genotype and seven pigmentation traits using a Kruskal–Wallis nonparametric test Significant associations (at the P < 0.05 level) are shown in bold Table Estimates of linkage disequilibrium (LD) for all pairwise comparisons between Mc1r SNP and four Agouti SNPs Both r Trait Mc1r Agouti Agouti Agouti Agouti Mc1r kb 40 kb 50 kb 100 kb and Agouti 40 kb∗ Rostrum Whisker Cheek patch Dorsal stripe Flank Ventrum Tail stripe 5.71 1.05 5.40 11.15 9.41 13.38 18.26 values (below the diagonal) and D’ (above the diagonal) are given Significant estimates of LD are shown in bold (P < 0.05) and were generated from 10,000 permutations Mc1r Agouti kb Agouti 40 kb Agouti 50 kb Agouti 100 kb Mc1r Agouti kb Agouti 40 kb Agouti 50 kb Agouti 100 kb – 0.002 0.021 0.005 0.001 0.063 – 0.014 0.004 0.002 0.425 0.160 – 0.028 0.006 0.335 1.000 0.675 – 0.002 0.121 0.352 0.146 1.000 – ∗ 0.80 0.24 1.17 3.35 1.51 6.36 7.61 17.07 23.30 41.32 22.28 18.51 15.35 26.07 1.02 3.73 0.41 0.25 0.16 0.11 0.23 4.50 7.32 5.98 2.76 2.74 1.97 3.05 Nine genotypic classes based on the genotypes at each locus EVOLUTION JULY 2008 1565 20.78 24.71 44.17 33.26 29.22 29.11 41.40 L M MULLEN AND H E HOEKSTRA tail stripe) For Agouti, we found no association between either the 50 kb or 100 kb SNP and any pigmentation trait The Agouti kb SNP showed an association with only ventrum pigment Most notably, we found the Agouti 40 kb SNP was significantly associated with all seven pigmentation traits Moreover, when genotypes at both Mc1r and the Agouti 40 kb SNP were considered together, there was also a significant association between genotype and all seven pigmentation traits—this was not true for any of the other three Agouti SNPs Discussion Using molecular genetics and modern mathematical analyses, we show that natural selection is almost certainly responsible for maintaining variation in pigmentation among the P polionotus populations first studied by Francis Sumner nearly a century ago Despite an abrupt change in pigmentation traits along the transect, populations experience high levels of gene flow, an observation inconsistent with the hypothesis that the phenotypic cline reflects recent secondary contact between populations that had diverged in allopatry Our results instead imply that the distribution of pigment variants is driven by strong environmentally based selection Because the mutations responsible for this pigment variation must also show clinal variation, we were able to identify an upstream regulatory region of one candidate pigmentation gene, Agouti, that may contribute to this classic phenotypic cline PHENOTYPE AND ENVIRONMENT Several morphological traits, including pigmentation, differ among populations in Sumner’s transect Foot length and tail length were measurably larger in mice inhabiting the sandy beaches than in those inhabiting oldfields These differences may have evolved in response to differences in soil type: larger feet and tails may provide greater stability for mice living on loose beach sand (Sumner 1917; Blair 1950; Hayne 1950) However, it remains unclear what proportion of the variation is explained by genetic versus environmental effects: it is possible, although unlikely, that foot and tail length are developmental rather than evolutionary responses to particular substrates (e.g., Losos et al 2000) The most striking phenotypic variation, however, is in pigmentation—with the lightest mice inhabiting white sand beaches Although a recent study has attributed pigment variation in beach mice solely to genetic drift (Van Zant and Wooten 2007), our data, combined with previous studies, strongly suggest that selection for crypsis is the primary mechanism driving the evolution of these interpopulation differences in pigmentation Laboratory crosses clearly show that pigmentation differences among Sumner’s populations are genetically based (Sumner 1930; Bowen and Dawson 1977) Enclosure experiments also showed that owls 1566 EVOLUTION JULY 2008 prey on Peromyscus that match their background substrate less often than more conspicuous mice (Dice 1947; Kaufman 1974a,b) Finally, the significant positive correlations between soil color and all seven pigmentation traits are consistent with selection for camouflage Nevertheless, the correlation between coat color and soil color is not perfect: contemporary soil samples decrease in brightness by more than 50% within the first 10 km from the coast, whereas most of the variation in mouse pigmentation occurs roughly 55 km from the coast There are several explanations for this discordance between the clines in soil color and pigmentation traits First (although this seems unlikely), increased urbanization and changes in land use could have altered soil conditions in the time since Sumner’s study, so that the break in soil color has recently shifted southward Asymmetric gene flow can also account for this discordance: light mice might, for example, disperse farther than dark ones However, our genetic data provide no evidence for asymmetric gene flow between populations Asymmetric selection may also contribute to the discordance between soil color and pigmentation For example, if predators are more efficient at capturing mice on different backgrounds, this could shift the break in pigmentation northward from the break in soil color, as is observed However, previous ecological-genetic studies suggest that selection may in fact be stronger against light mice on dark substrate than vice versa (Hoekstra et al 2004), a result consistent with predation experiments on Peromyscus (Dice 1947) A related and more plausible explanation is that additional environmental factors (in combination with soil color) impose differential selection on populations The type and density of vegetative cover, for instance, may also influence predation risks Although beach populations are largely devoid of vegetation, more northern populations have denser and more diverse vegetation Therefore, selection could be more intense on all mice on the beach compared to more northern environments (thus selection against dark mice on beaches would be stronger than light mice in the mainland), a scenario that could produce the observed discordance For now, however, this discordance between environment and pigmentation remains untested and largely unexplained Although we believe that selection is the primary mechanism of coat color variation among populations, there is an alternative explanation: the clinal variation resulted from secondary contact between formerly isolated populations that diverged in pigmentation while allopatric Bowen (1968) noted that the area of abrupt soil change coincided with the highest Pleistocene shoreline, so that the decrease in water levels could have allowed a divergent (and light-colored) subspecies to invade the more coastal habitat However, under the scenario of recent secondary contact, we expect that all characters (both genetic and phenotypic) would share a geographic break at the same location For phenotypic traits, differentiation in foot size and tail length occurs more than C L I NA L S E L E C T I O N I N P E RO M Y S C U S 25 km south of where the change in brightness occurs, a pattern inconsistent with this expectation And for neutral genetic markers, we see very little genetic differentiation among populations Most striking is the complete lack of genetic structure among populations surrounding the Intergrades (IN), where most of the change in pigmentation occurs Thus, we see no evidence for recent admixture of diverged populations, or any contemporary barriers to gene flow Both phenotypic data and neutral genetic markers suggest that clinal variation in pigmentation is best explained by strong habitat-specific selection These data, however, not rule out the possibility of older secondary contact between divergent forms and subsequently the maintenance of pigmentation differences by habitatspecific selection Careful phylogeographic studies may be able to provide information about the geographic origins of and divergence times between the terminal populations in this cline However, even if secondary contact was ancient, we must still invoke selection to explain the disparity between the geographic differentiation in color and the geographic uniformity of many genetic markers In either case, selection must be strong to maintain habitatspecific phenotypes in the face of high levels of gene flow In fact, Haldane (1948) estimated that selection coefficients were around 10 percent Using statistical cline-fitting methods, we indeed found that large selection coefficients are necessary to explain the observed distribution of pigmentation traits Our estimates calculated with the ecotone model were remarkably similar in magnitude to those estimated by Haldane However, a model assuming a spatial selection gradient also may be appropriate, because the phenotypic cline widths (> km) are slightly larger than the estimated dispersal distance of the mice (< km) As expected, estimates of selection calculated with the gradient model were lower than those from the ecotone model It is also important to note that mark–recapture studies often underestimate true dispersal distance (Mallet et al 1990; McCallum 2000), thus, our selection estimates may be conservative Therefore, it is likely that the true selection estimates fall between these two extremes Under both models, our estimates of selection coefficients are especially high for several traits—whisker, cheek, flank, and dorsal stripe—perhaps because these traits are the most visible to predators Less-visible traits, such as the ventrum, show the smallest estimates of selection However, because all pigmentation traits are phenotypically and genetically correlated (Hoekstra et al 2006; Steiner et al 2007), it is unclear which is or are the direct target(s) of selection ADAPTIVE GENETIC VARIATION Previous studies identified two genes, Mc1r and Agouti, that explain most of the variation in pigment between P polionotus leucocephalus, the light-colored Santa Rosa Island beach mouse, and P p subgriseus, a dark mainland mouse (Hoekstra et al 2006; Steiner et al 2007) Like leucocephalus, the beach population in Sumner’s transect (P p allophrys) is significantly lighter than any mainland subspecies However, there are some differences in coloration between beach subspecies: allophyrs has more pigmented regions and is slightly darker than leucocephalus Despite these phenotypic differences, we were surprised that there was no clinal variation in the frequency of Mc1r alleles Instead, one puzzling result emerged: we found the light Mc1r allele in dark-colored mice (e.g., mice from the northernmost populations) despite previous experiments showing that the light Mc1r allele causes light pigmentation (Hoekstra et al 2006) One explanation is that the effects of the light Mc1r allele in Sumner’s populations could be different from that in the populations where it was first identified (i.e., P p leucocephalus) due to differences in genetic background For example, previous work showed that the light Mc1r mutation does not significantly decrease pigmentation if found on a dark Agouti background (Steiner et al 2007) Thus, we hypothesize the dark mice in the northernmost populations that harbor light Mc1r alleles also have dark Agouti alleles It is also possible, of course, that other pigmentation genes could be responsible for the similar light color phenotype despite similar selective pressures Unlike Mc1r, we did find clinal allelic variation at the Agouti locus Two SNPs, 40 and 100 kb, varied significantly with geographic distance This suggests that either two different mutations in the Agouti cis-regulatory region contribute to pigmentation, or that both SNPs are in LD with the same causal mutation(s) Additional genotyping at markers between the 40 kb and 100 kb SNPs will distinguish between these explanations Of these two SNPs, the Agouti 40 kb SNP may be closest to a causal mutation for several reasons First, the light allele is fixed in the light-colored southern populations and occurs at lower frequency, although is not absent, in dark northern populations This suggests that the 40 kb SNP may be in LD with a causal SNP that is driving the clinal variation, but that the SNP is still some distance away from it, leading to an imperfect association Second, the center and width of cline in allele frequency for the 40 kb SNP is statistically indistinguishable from those of several pigmentation traits Third, we found a statistical association between genotypes at this locus and all seven pigmentation phenotypes Fourth, the 40 kb SNP is in LD with the Mc1r SNP, which is physically located on another chromosome, and this pattern is consistent with expectations based on their known epistatic interaction (Steiner et al 2007) This LD reflects a disproportionately high association of the light Mc1r allele with the dark Agouti allele in the northernmost populations, as predicted if these mice are darkcolored Moreover, when genotypes at both Mc1r and the Agouti 40 kb SNP are considered together, there is a statistically significant association between genotype and all seven pigmentation EVOLUTION JULY 2008 1567 L M MULLEN AND H E HOEKSTRA traits Additional genotyping at markers around this 40 kb SNP will help narrow the region with the goal of eventually identifying the causal mutation(s) in this region These results highlight the potential use of natural populations, and clinal variation in particular, for fine-scale mapping of mutations (Stinchcombe and Hoekstra 2008) Together, these data raise several possibilities about the genetic bases of Sumner’s classic pigmentation cline First, either this Agouti 40 kb mutation—or, more likely, one or more mutations in LD with this mutation—may contribute to pigmentation differences among populations Second, it is likely that this mutation(s) interacts epistatically with the Mc1r mutation, producing the LD observed between these loci Third, previous studies suggest that pigment variation in these populations results from the action of several genes—Sumner’s (1930) genetic analyses of this transect and Bowen and Dawson’s (1977) study of pigment pattern variation in natural populations each independently estimated that approximately 10 genes underlie differences between beach and mainland forms Our approach also may be valuable in identifying any additional genes that contribute to naturally occurring variation in pigmentation (or other traits) Together, these data show that natural selection acting on phenotypes produces allele frequency clines in the underlying genes, thus allowing us to narrow in on the precise mutations responsible for adaptive traits ACKNOWLEDGMENTS We thank several colleagues: J Patton for access to Sumner’s specimens, T Belfiore for DNA extraction protocols, S Hoang for field assistance, M Streisfeld for help with spectrophotometric analysis, and S Ramdeen for providing soil samples from the Florida Geological Survey Members of the Hoekstra laboratory, J Coyne, J Huelsenbeck, J Kohn, P Morin, K Ruegg, and E Wilson provided helpful discussion, and additional comments from B Blackman, D Haig, A Kaliszewska, M Patten, J Stinchcombe, and two anonymous reviewers greatly improved this manuscript This work was supported by an American Society of Mammalogists Grant-in-Aid, a NSF Graduate Research Fellowship and a NSF Dissertation Improvement Grant to LMM, and NSF grant (DEB-0344710) to HEH LITERATURE CITED Barton, N H., and S J E Baird 1995 Analyse: an application for analysing hybrid zones Available at http://www.biology.ed.ac.uk/ research/institutes/evolution/software.php, Edinburgh Barton, N H., and K S Gale 1993 Genetic analysis of hybrid zones Pp 13–45 in R G Harrison, ed Hybrid zones and the evolutionary process Oxford Univ Press, Oxford, UK Bazykin, A D 1969 Hypothetical mechanism of speciation Evolution 23:685–687 Beerli, P., and J Felsenstein 1999 Maximum-likelihood estimation of migration rates and effective population numbers in two populations using a coalescent approach Genetics 152:763–773 Bennett, A T D., and I C Cuthill 1994 Ultraviolet vision in birds: what is its function? Vision Res 34:1471–1478 1568 EVOLUTION JULY 2008 Berry, A., and M Kreitman 1993 Molecular analysis of an allozyme cline: alcohol dehydrogenase in Drosophila melanogaster on the east coast of North America Genetics 134:869–893 Blair, W F 1946 An estimate of the total number of beach-mice of the subspecies Peromyscus polionotus leucocephalus, occupying Santa Rosa Island, Florida Am Nat 80:665–668 ——— 1950 Ecological factors in speciation of Peromyscus Evolution 4:253–275 ——— 1951 Population structure, social behavior, and environmental relations in a natural population of the beach mouse (Peromyscus polionotus leucocephalus) Contrib Lab Vertebrate Biol Univ Michigan 48:1– 47 Bonin, A., E Bellemain, P B Eidesen, F Pompanon, C Brochman, and P Taberlet 2004 How to track and assess genotyping errors in population genetics studies Mol Ecol 13:3261–3273 Bowen, W W 1968 Variation and evolution of gulf coast populations of beach mice, Peromyscus polionotus Bulletin of the Florida State Museum 12:1–91 Bowen, W W., and W D Dawson 1977 Genetic analysis of coat pattern variation in oldfield mice (Peromyscus polionotus) of western Florida J Mammal 58:521–530 Bradbury, P J., Z Zhang, D E Kroon, T M Casstevens, Y Ramdoss, and E S Buckler 2007 TASSEL: software for association mapping of complex traits in diverse samples Bioinformatics 23:2633–2635 Caicedo, A L., J R Stinchcombe, K M Olsen, J Schmitt, and M D Purugganan 2004 Epistatic interaction between Arabidopsis FRI and FLC flowering time genes generates a latitudinal cline in a life history trait Proc Natl Acad Sci USA 101:15670–15675 Casgrain, P., and P Legendre 2001 The R package for multivariate and spatial analysis Universite de Montreal, Departement de Sciences Biologiques Dice, L R 1947 Effectiveness of selection by owls of deermice (Peromyscus maniculatus) which contrast in color with their background Contrib Lab Vertebrate Biol Univ Michigan 34:1–20 Dobzhansky, T 1937 Genetics and the origin of species Columbia Univ Press, New York Eanes, W F 1999 Analysis of selection on enzyme polymorphisms Annu Rev Ecol Syst 30:301–326 Edwards, A 1992 Likelihood Johns Hopkins Univ Press, Baltimore, MD Endler, J A 1977 Geographic variation, speciation, and clines Princeton Univ Press, Princeton, NJ ——— 1990 On the measurement and classification of color in studies of animal color patterns Biol J Linn Soc 41:315–352 Evanno, G., S Regnaut, and J Goudet 2005 Detecting the number of clusters of individuals using the software STRUCTURE: a simulation study Mol Ecol 14:2611–2620 Ford, E B 1954 Problems in the evolution of geographic races P 105 in J Huxley, A C Hardy, and E B Ford, eds Evolution as a process Allen and Unwin, London ——— 1960 Evolution in process Pp 188–189 in S Tax, ed Evolution after Darwin Univ of Chicago, Chicago ——— 1964 Ecological genetics Chapman and Hall, London Fry, J D., K Donlon, and M Saqeikis 2008 A worldwide polymorphism in Aldehyde dehydrogenase in Drosophila melanogaster: evidence for selection mediated by dietary ethanol Evolution 62:66–75 Haldane, J B S 1948 The theory of a cline J Genet 48:277–284 Hayne, D 1950 Reliability of laboratory-bred stocks as samples of wild populations, as shown in a study of the variation of Peromyscus polionotus in parts of Florida and Alabama Contrib Lab Vertebrate Biol Univ Michigan 46:1–53 Hedrick, P W 1987 Gametic disequilibrium measures: proceed with caution Genetics 117:331–341 C L I NA L S E L E C T I O N I N P E RO M Y S C U S ——— 2006 Genetic polymorphism in heterogeneous environments: the age of genomics Annu Rev Ecol Evol Syst 37:67–93 Hoekstra, H E., K E Drumm, and M W Nachman 2004 Ecological genetics of adaptive color polymorphism in pocket mice: geographic variation in selected and neutral genes Evolution 58:1329–1341 Hoekstra, H E., R J Hirschmann, R A Bundey, P A Insel, and J P Crossland 2006 A single amino acid mutation contributes to adaptive beach mouse color pattern Science 313:101–4 Howell, A H 1920 Description of a new species of beach mouse in Florida J Mammal 1:237–240 Huxley, J 1943 Evolution: the modern synthesis Harper and Brothers, New York Kaufman, D W 1974a Adaptive coloration in Peromyscus polionotus: experimental selection by owls J Mammal 55:271–283 ——— 1974b Differential owl predation on white and agouti Mus musculus Auk 91:145–150 Losos, J B., D A Creer, D Glossip, R Goellner, A Hampton, G Roberts, N Haskell, P Taylor, and J Etting 2000 Evolutionary implications of phenotypic plasticity in the hindlimb of the lizard Anolis sagrei Evolution 54:301–305 Mallet, J., N Barton, G Lamas M., J Santisteban C., M Muedas, and H Eeley 1990 Estimates of selection and gene flow from measures of cline width and linkage disequilibrium in Heliconius hybrid zones Genetics 124:921–936 Mayr, E 1942 Systematics and the origin of species Columbia Univ Press, New York City ——— 1954 Evolution as a process P 105 in J Huxley, A C Hardy, and E B Ford, eds Evolution as a process Allen and Unwin, London ——— 1963 Animal species and evolution Belknap Press of Hrvard Univ Press, Cambridge, MA McCallum, H 2000 Population parameters: estimation for ecological models Blackwell Publishing, Oxford Morin, P A., K E Chambers, C Boesch, and L Vigilant 2001 Quantitative polymerase chain reaction analysis of DNA from noninvasive samples for accurate microsatellite genotyping of wild chimpanzees (Pan troglodytes verus) Mol Ecol 10:1835–1844 Mullen, L M., R L Hirschmann, K L Prince, T L Glenn, M J Dewey, and H E Hoekstra 2006 Sixty polymorphic microsatellite markers for the oldfield mouse developed in Peromyscus polionotus and P maniculatus Mol Ecol Notes 6:36–40 Peakall, R., and P E Smouse 2006 GenALEX 6: genetic analysis in Excel Population genetic software for teaching and research Mol Ecol Notes 6:288–295 Phillips, B L., S J E Baird, and C Moritz 2004 When vicars meet: a narrow contact zone between morphologically cryptic phylogeographic lineages of the rainforest skink, Carlia rubrigularis Evolution 58:1536– 1548 Pritchard, J K., M Stephens, and P Donnelly 2000 Inference of population structure using multilocus genotype data Genetics 155:945–959 Raymond, M., and F Rousset 1995 GENEPOP (version 3.4): population genetics software for exact tests and ecumenicism J Heredity 86:248– 249 Schemske, D W., and P Bierzychudek 2007 Spatial differentiation for flower color in the desert annual Linanthus parryae: was Wright right? Evolution 61:2528–43 Schmidt, P S., and D M Rand 2001 Adaptive maintenance of genetic polymorphism in an intertidal barnacle: habitat- and life-stage-specific survivorship of MPI genotypes Evolution 55:1336–1344 Schneider, S., D Roessli, and L Excoffier 2000 Arlequin ver 2.000: A software for population genetics data analysis Univ of Geneva, Genetics Biometry Laboratory, Switzerland Slatkin, M 1973 Gene flow and selection in a cline Genetics 75:733–56 ——— 1995 A measure of population subdivision based on microsatellite allele frequencies Genetics 139:457–462 Steiner, C C., J N Weber, and H E Hoekstra 2007 Adaptive variation in beach mice produced by two interacting pigmentation genes PLoS Biol 5:e219 Stinchcombe, J R., and H E Hoekstra 2008 Population genomics and quantitative genetics: finding the genes underlying ecologically important traits Heredity 100:158–170 Sumner, F B 1917 The role of isolation in the formation of a narrowly localized race of deer-mice (Peromyscus) Am Nat 51:173–185 ——— 1926 An analysis of geographic variation in mice of the Peromyscus polionotus group from Florida and Alabama J Mammal 7:149–184 ——— 1929a The analysis of a concrete case of intergradation between two subspecies Proc Natl Acad Sci USA 15:110–120 ——— 1929b The analysis of a concrete case of intergradation between two subspecies II additional data and interpretations Proc Natl Acad Sci USA 15:481–493 ——— 1930 Genetic and distributional studies of three subspecies of Peromyscus J Genet 23:275–376 Swilling, J W R., and M C Wooten 2002 Subadult dispersal in a monogamous species: the Alabama Beach Mouse (Peromyscus polionotus ammobates) J Mammal 83:252–259 Szymura, J M., and N H Barton 1986 Genetic analysis of a hybrid zone between the fire-bellied toads, Bombina bombina and B variegata, near Cracow in southern Poland Evolution 40:1141–1159 Taberlet, P., S Friffin, B Goossens, S Questiau, V Manceau, N Escaravage, L Waits, and J Bouvet 1996 Reliable genotyping of samples with very low DNA quantities using PCR Nucleic Acids Res 24:3189–3194 Van Zant, J L., and M C Wooten 2007 Old mice, young islands and competing biogeography hypotheses Mol Ecol 16:5070–5083 Associate Editor: J Feder EVOLUTION JULY 2008 1569 L M MULLEN AND H E HOEKSTRA Supplementary Material The following supplementary material is available for this article: Table S1 Table S2 Table S3 Table S4 Table S5 Sample size and collection numbers for each population Morphological variation among populations Pelage and soil reflectance measurements for each population Forward and reverse primer and TaqMan probe sequences for three Agouti SNPs Summary of genetic diversity for each population This material is available as part of the online article from: http://www.blackwell-synergy.com/doi/abs/10.1111/j.1558-5646.2008.00425.x (This link will take you to the article abstract) Please note: Blackwell Publishing is not responsible for the content or functionality of any supplementary materials supplied by the authors Any queries (other than missing material) should be directed to the corresponding author for the article 1570 EVOLUTION JULY 2008 [...]... genetic analyses of this transect and Bowen and Dawson’s (1977) study of pigment pattern variation in natural populations each independently estimated that approximately 10 genes underlie differences between beach and mainland forms Our approach also may be valuable in identifying any additional genes that contribute to naturally occurring variation in pigmentation (or other traits) Together, these data... Society of Mammalogists Grant -in- Aid, a NSF Graduate Research Fellowship and a NSF Dissertation Improvement Grant to LMM, and NSF grant (DEB-0344710) to HEH LITERATURE CITED Barton, N H., and S J E Baird 1995 Analyse: an application for analysing hybrid zones Available at http://www.biology.ed.ac.uk/ research/institutes/evolution/software.php, Edinburgh Barton, N H., and K S Gale 1993 Genetic analysis of... (Peromyscus) Am Nat 51:173–185 ——— 1926 An analysis of geographic variation in mice of the Peromyscus polionotus group from Florida and Alabama J Mammal 7:149–184 ——— 192 9a The analysis of a concrete case of intergradation between two subspecies Proc Natl Acad Sci USA 15:110–120 ——— 1929b The analysis of a concrete case of intergradation between two subspecies II additional data and interpretations Proc Natl Acad... Zhang, D E Kroon, T M Casstevens, Y Ramdoss, and E S Buckler 2007 TASSEL: software for association mapping of complex traits in diverse samples Bioinformatics 23:2633–2635 Caicedo, A L., J R Stinchcombe, K M Olsen, J Schmitt, and M D Purugganan 2004 Epistatic interaction between Arabidopsis FRI and FLC flowering time genes generates a latitudinal cline in a life history trait Proc Natl Acad Sci USA... A C Hardy, and E B Ford, eds Evolution as a process Allen and Unwin, London ——— 1963 Animal species and evolution Belknap Press of Hrvard Univ Press, Cambridge, MA McCallum, H 2000 Population parameters: estimation for ecological models Blackwell Publishing, Oxford Morin, P A. , K E Chambers, C Boesch, and L Vigilant 2001 Quantitative polymerase chain reaction analysis of DNA from noninvasive samples... leucocephalus, the light-colored Santa Rosa Island beach mouse, and P p subgriseus, a dark mainland mouse (Hoekstra et al 2006; Steiner et al 2007) Like leucocephalus, the beach population in Sumner’s transect (P p allophrys) is significantly lighter than any mainland subspecies However, there are some differences in coloration between beach subspecies: allophyrs has more pigmented regions and is slightly... MULLEN AND H E HOEKSTRA Supplementary Material The following supplementary material is available for this article: Table S1 Table S2 Table S3 Table S4 Table S5 Sample size and collection numbers for each population Morphological variation among populations Pelage and soil reflectance measurements for each population Forward and reverse primer and TaqMan probe sequences for three Agouti SNPs Summary of... morphological traits, including pigmentation, differ among populations in Sumner’s transect Foot length and tail length were measurably larger in mice inhabiting the sandy beaches than in those inhabiting oldfields These differences may have evolved in response to differences in soil type: larger feet and tails may provide greater stability for mice living on loose beach sand (Sumner 1917; Blair 1950; Hayne 1950)... for maintaining variation in pigmentation among the P polionotus populations first studied by Francis Sumner nearly a century ago Despite an abrupt change in pigmentation traits along the transect, populations experience high levels of gene flow, an observation inconsistent with the hypothesis that the phenotypic cline reflects recent secondary contact between populations that had diverged in allopatry... ventrum and Mc1r AND AGOUTI SNPs All pairwise estimates of LD between Mc1r and Agouti SNPs are reported in Table 5 r2 values ranged from 0.001 to 0.028, Table 6 Association (χ2 values) between genotype and seven pigmentation traits using a Kruskal–Wallis nonparametric test Significant associations (at the P < 0.05 level) are shown in bold Table 5 Estimates of linkage disequilibrium (LD) for all pairwise